The development of Single Point Diamond Turning (SPDT) technique provided technologic and technical base for the mass production and application of aluminum mirrors. To ensure the image quality of the optical system, stress-caused distortion of reflective surface by mounting pretension stress and uneven mounting surfaces needed to be attenuated. Multiple typical structural shapes of the stress-relief support were demonstrated and the design principle was summarized. The SPDT technique created a new technical route to aid the optical assembly process. The design of assembly assisted by SPDT was showed in instances of reflective surfaces on unilateral side and bilateral sides in single mirror and detached double mirrors. The integral SPDT machining flow of optical surfaces, structural mounting surfaces, transferring datum and reference surfaces of interference detection linking with concrete examples implemented the transfer of high-precision assembly reference and ensured high-accuracy alignment of mirrors. This method could improve the assembly efficiency and shorten the manufacture period of optical systems.

The manufacture of high aspect-ratio hard X-ray zone plates for Inertial Confinement Fusion (ICF) experiment by using electron beam lithography and X-ray lithography was demonstrated. The X-ray lithography mask was first fabricated on hollow polyimide film to reduce the back-scattering during electron-beam lithography process. Then multiple X-ray exposures and full water electroplating method were used to increase the lodging resistance of the resist pattern. The zone plates with outermost zone width of 350 nm, profile thickness of 3.5 μm, and aspect ratio of 10, were successfully fabricated by using the combined technology. The structure of zone plates is steep and has excellent graphics quality, and can be used for 10 to 30 keV X-ray imaging system.

For unreasonable attention model, limitations of optical estimation and noise, the result of existing computational method of visual motion attention sometimes can not reflect real motion attention information for video scene of motion complex. A new computational method of visual motion attention is suggested. In the method, motion attention model is firstly built according to the mechanism of attention. Then, the visual motion attention is computed on basis of variable size block motion estimation. Finally, the final salience map come from the fusion of multi-scale motion attention salience map. The experimental results also show the calculation result of this method is better than that of the existing method.

A portable scanning surface plasmon resonance biosensor was constructed for the safety detection of food. The optical system, hardware circuit and software design were thoroughly introduced. The clenbuterol hydrochloride was used as samples, and the biological probe was fixed on the sensor chip. Clenbuterol hydrochloride antibody with different concentration as 2 mg/L, 4 mg/L, 8 mg/L,16 mg/L, and 32 mg/L were continuously detected. The experimental results accord with the immune response rule of biological molecules. The validity of this method and device was demonstrated by experiments. The device is portable and easy to operate. The device has extensive prospect in the safety detection of food.

Reliable and efficient Optic Disc (OD) localization is significant to automatic retinal image diagnosis. A fast automatic OD localization algorithm is presented. First, candidate OD locations are identified by gradient and intensity information. Then, by observation, the main vasculature are converged in OD region and spread along vertical direction. A template is used to locate the true OD center. The algorithm was tested on five publicly-available databases. The OD was successfully located in 1 514 images out of 1 540 images (98.3%). The averaged time for one STARE image is only 6.3 second. And the proposed method shows good robustness in both normal and diseased images.

In order to realize close-range measurement, a miniaturized bionic compound-eye system was developed and the design principles of the system structure and parameters were introduced. Besides, the calibration and three-dimensional measurement algorithms were investigated. First, the platform for calibration and measurement was built based on imaging characteristics of the compound-eye. Three kinds of camera calibration methods, including Zhang’s method, the direct linear transformation method and Tsai two-step method, were carried out to calibrate the central sub-eye. By comparing the experimental results, it was determined that Tsai two-step method with high precision and accuracy was suitable for the calibration of the compound-eye system presented. Then, to solve the problem that the edge sub-vision axis was not perpendicular to the image sensor, a new distortion model that effectively improved the edge sub-eye calibration accuracy was proposed. Finally, a multiple sub-eye detection model was established, and its impact on the compound-eye camera measurement accuracy were discussed, from which it was determined that three sub-eye condition had better accuracy and stability than the two-eye. Experimental results indicate that the relative error of the compound eye three-dimensional detection system is about 2% when the object distance ranges from 150～260 mm, and this system could achieve the instrument miniaturization and basic close-range measurement.

In order to meet the needs of intelligent video surveillance, an unsupervised abnormal detecting algorithm was proposed. Firstly, model of mixture of Gaussians was used to extract the motion area, and the motion area was labeled. Then, observation sequence updated in real-time of feature matrix was established by the optical flow features obtained from labeled area which was normalized to the feature matrix. Finally, applying reconstruction works of two-dimensional principal component analysis on the sequence, abnormal behavior can be detected according to the energy ratio between the recovered feature matrix and original feature matrix. Experiments were conducted on various video datasets, which shows the effectiveness of the proposed method.

The recording system used in aviation, requires a higher maintainability and reliability compared with ground system. The volume and dissipation also have strict limit. In order to solve the requests, embedded NAND Flash image recording system based on Austin Trumpbell Radio (ATR) reinforced structure is proposed. The system uses CPCIE 3U standard for design and is placed in the 1/2 ATR box, so the system has a higher maintainability and reliability. High-speed SRIO bus and NAND Flash recording arrays are used to achieve a high-speed real-time storage. The paper introduces the design of every card. The test results show the system run stably and reliably .The recording speed of the system is up to 912 MB/s and satisfies the recording requirement for high-resolution aviation CCD camera.

Different imaging mechanism of Synthetic Aperture Radar (SAR) and visible images in the general fusion methods make it difficult to achieve good fusion effect. For the target recognition, firstly get fusion image by Nonsubsampled Contourlet Transform (NSCT) to fuse the important objective information of SAR image into visible image, and retain detail information as much as possible. Then, combine mathematical morphology with multiscale space theory to get the bright and dark detail features of the original images. Finally, obtain fused image with significantly enhanced bright and dark detail features. Experimental result shows that the proposed algorithm can effectively fuse target information of SAR image, enhance the detail features of the source image, achieve a better visual effect, and improve the capability of target detection and identification.

Local invariant algorithms are usually not applicable for object detection with variance of shape and surface texture, and existing local descriptors show limited ability to distinguish this kind of shape. A new object detection algorithm based on orientation descriptor is proposed. Initial descriptors are calculated based on silhouette of model or edge image of testing image, on this basis of which, orientation descriptor are calculated for each pixel in image. Orientation descriptor can describe edge orientation and tolerate small shape deformation. Multi-resolution method is utilized to speed up sliding-window algorithm. Descriptor matrixes of valid candidate region and model are matched to determine the existence of object at this position. The experimental results show that the proposed algorithm achieves fine detection rate relatively.

Fourier descriptor is an important classification method of object contours. At present, most of researchers use only the amplitude of Fourier descriptor whose accuracy is not enough. Aiming at the problem, we presented three methods of fusion of Fourier descriptor’s phase and amplitude to classification of object contours and compared their results. Experimental results show that pixel-level fusion is the most effective method whose classification accuracy is 10% higher than the method for only using the amplitude of Fourier descriptor, but the pixel-level fusion method is timeconsuming because its number of feature vector’s dimensions is much. Comparing Fourier descriptors of pixel level fusion with curvature scale space descriptors and shape context, it is seen that Fourier descriptors are higher timeliness.

The registration of satellite imagery is challenging task because of considerable differences between the image pairs captured by different sensors, view angles, spectrum or at different times. To align the images effectively, a coarse-to-fine registration algorithm is proposed. First, Fourier-Mellin transform algorithm is used to implement a fast coarse registration. Then, the optimization algorithm based on MSSIM measure is used to implement a fine registration. Because there is no accurate measure standard, it is very difficult to give a quantitative assessment of the registration results for real satellite images. A novel registration evaluation method is proposed, called Catching Curve Feature Evaluation (MCfe) method. In MCfe, kurtosis, peak deviation, peak maximum and Root Mean Square Error (RMSE) among peak maxima are extracted as the quantitative evaluation indexes, and the minimizing on RMSE is used as a registration criterion. The results demonstrate that the coarse-to-fine registration algorithm can achieve an extremely accurate registration for satellite imagery. The MCfe method can intuitively describe the registration performance from the curve features such as smoothness and sharpness, and also to be used to quantitatively assess the registration results, but also to automatically adjust the registration parameters to obtain a more accurate registration.

Fusion multi-channel satellite cloud images can provide a more comprehensive and reliable information for monitoring and forecasting weather conditions. A fusion method for satellite cloud images using Aliasing-suppression and Shift-invariance Contourlet Transform (AS&SICT) is presented. Firstly, in order to deal with the frequency aliasing and shift-variance problems of the original Contourlet transform, the aliasing-free filter bank is combined with non-subsampled directional filer banks (NDFB) to construct AS & SICT. Then, two channels of satellite images (infrared and visible) are decomposed into low-frequency and high-frequency directional sub-bands by AS&SICT. Next, low-frequency sub-band coefficients are fused by weighted regional energy rule while high-frequency sub-bands coefficients are fused by weighted regional variance rule. Finally, we get the fused cloud image by taking the inverse AS&SICT on the fusion coefficients. Experimental results show that the proposed fusion method can not only improve the resolution of infrared cloud image by joining the texture details of visible cloud image, but also keep the original brightness temperature information of infrared cloud image better.

An online combined two kinds of features method for visual object tracking was proposed. A feature set was built by combining object contour’s linear parametric equation and edge histogram. Firstly, the Hough transform is used to detect line in picture and obtain the parameters of the line, meanwhile, get the histogram at the same position of the line. The resulting initial template contains the location and pixel distribution information of the target. Secondly, in the subsequent frames, pixels divided and clustered by gradient direction and parameter equation would form more than one line which was to be matched. At last, obtain the exact location of the straight edge using the histogram matching algorithm. The experimental results indicate that the algorithm can run effectively when occlusion and complicated backgrounds happened, and perform favorably on challenging sequences in terms of efficiency, accuracy and robustness.